ICML 2024 论文列表

Sharpness-aware minimization (SAM) methods have gained increasing popularity by formulating the problem of minimizing both loss value and loss sharpness as a minimax objective. In this work, we increase the efficiency of the maximization and minimization parts of SAM's objective to achieve a better loss-sharpness trade-off. By taking inspiration from the Lookahead optimizer, which uses multiple descent steps ahead, we propose Lookbehind, which performs multiple ascent steps behind to enhance the maximization step of SAM and find a worst-case perturbation with higher loss. Then, to mitigate the variance in the descent step arising from the gathered gradients across the multiple ascent steps, we employ linear interpolation to refine the minimization step. Lookbehind leads to a myriad of benefits across a variety of tasks. Particularly, we show increased generalization performance, greater robustness against noisy weights, as well as improved learning and less catastrophic forgetting in lifelong learning settings. Our code is available at https://github.com/chandar-lab/Lookbehind-SAM.

Understanding the interactions and interplay of microorganisms is a great challenge with many applications in medical and environmental settings. In this work, we model bacterial communities directly from their genomes using graph neural networks (GNNs). GNNs leverage the inductive bias induced by the set nature of bacteria, enforcing permutation invariance and granting combinatorial generalization. We propose to learn the dynamics implicitly by directly predicting community relative abundance profiles at steady state, thus escaping the need for growth curves. On two real-world datasets, we show for the first time generalization to unseen bacteria and different community structures. To investigate the prediction results more deeply, we create a simulation for flexible data generation and analyze effects of bacteria interaction strength, community size, and training data amount.

The widespread use of artificial intelligence requires finding energy-efficient paradigms for the field. We propose to reduce the energy consumption of Gaussian process regression using low-precision floating-point representations. We explore how low-precision representations impact the results of Gaussian process regression and how data set properties, implementation approach, model performance, and energy consumption interact. Our findings show that a well-conditioned kernel matrix allows reducing the energy consumption by up to 89.01% for 98.08% of arithmetic operations with little to no impact on model performance. Our findings are relevant whenever one needs to invert a symmetric full-rank matrix.

We investigate model assessment and selection in a changing environment, by synthesizing datasets from both the current time period and historical epochs. To tackle unknown and potentially arbitrary temporal distribution shift, we develop an adaptive rolling window approach to estimate the generalization error of a given model. This strategy also facilitates the comparison between any two candidate models by estimating the difference of their generalization errors. We further integrate pairwise comparisons into a single-elimination tournament, achieving near-optimal model selection from a collection of candidates. Theoretical analyses and empirical experiments underscore the adaptivity of our proposed methods to the non-stationarity in data.

In mobile and IoT systems, Federated Learning (FL) is increasingly important for effectively using data while maintaining user privacy. One key challenge in FL is managing statistical heterogeneity, such as non-i.i.d. data, arising from numerous clients and diverse data sources. This requires strategic cooperation, often with clients having similar characteristics. However, we are interested in a fundamental question: does achieving optimal cooperation necessarily entail cooperating with the most similar clients? Typically, significant model performance improvements are often realized not by partnering with the most similar models, but through leveraging complementary data. Our theoretical and empirical analyses suggest that optimal cooperation is achieved by enhancing complementarity in feature distribution while restricting the disparity in the correlation between features and targets. Accordingly, we introduce a novel framework, FedSaC, which balances similarity and complementarity in FL cooperation. Our framework aims to approximate an optimal cooperation network for each client by optimizing a weighted sum of model similarity and feature complementarity. The strength of FedSaC lies in its adaptability to various levels of data heterogeneity and multimodal scenarios. Our comprehensive unimodal and multimodal experiments demonstrate that FedSaC markedly surpasses other state-of-the-art FL methods.

We present an adaptive approach for robust learning from corrupted training sets. We identify corrupted and non-corrupted samples with latent Bernoulli variables and thus formulate the learning problem as maximization of the likelihood where latent variables are marginalized. The resulting problem is solved via variational inference, using an efficient Expectation-Maximization based method. The proposed approach improves over the state-of-the-art by automatically inferring the corruption level, while adding minimal computational overhead. We demonstrate our robust learning method and its parameter-free nature on a wide variety of machine learning tasks including online learning and deep learning where it adapts to different levels of noise and maintains high prediction accuracy.

Bridging the gap between diffusion models and human preferences is crucial for their integration into practical generative workflows. While optimizing downstream reward models has emerged as a promising alignment strategy, concerns arise regarding the risk of excessive optimization with learned reward models, which potentially compromises ground-truth performance. In this work, we confront the reward overoptimization problem in diffusion model alignment through the lenses of both inductive and primacy biases. We first identify a mismatch between current methods and the temporal inductive bias inherent in the multi-step denoising process of diffusion models, as a potential source of reward overoptimization. Then, we surprisingly discover that dormant neurons in our critic model act as a regularization against reward overoptimization while active neurons reflect primacy bias. Motivated by these observations, we propose Temporal Diffusion Policy Optimization with critic active neuron Reset (TDPO-R), a policy gradient algorithm that exploits the temporal inductive bias of diffusion models and mitigates the primacy bias stemming from active neurons. Empirical results demonstrate the superior efficacy of our methods in mitigating reward overoptimization. Code is avaliable at https://github.com/ZiyiZhang27/tdpo.

Partial differential equations (PDEs) are instrumental for modeling dynamical systems in science and engineering. The advent of neural networks has initiated a significant shift in tackling these complexities though challenges in accuracy persist, especially for initial value problems. In this paper, we introduce the *Time-Evolving Natural Gradient (TENG)*, generalizing time-dependent variational principles and optimization-based time integration, leveraging natural gradient optimization to obtain high accuracy in neural-network-based PDE solutions. Our comprehensive development includes algorithms like TENG-Euler and its high-order variants, such as TENG-Heun, tailored for enhanced precision and efficiency. TENG's effectiveness is further validated through its performance, surpassing current leading methods and achieving *machine precision* in step-by-step optimizations across a spectrum of PDEs, including the heat equation, Allen-Cahn equation, and Burgers' equation.

Many existing two-phase kernel-based hypothesis transfer learning algorithms employ the same kernel regularization across phases and rely on the known smoothness of functions to obtain optimality. Therefore, they fail to adapt to the varying and unknown smoothness between the target/source and their offset. This paper introduces Smoothness Adaptive Transfer Learning (SATL), a two-phase kernel ridge regression (KRR)-based algorithm to address these limitations. We first demonstrate that employing a misspecified fixed bandwidth Gaussian kernel in target-only KRR learning can achieve minimax optimality when the true function resides in Sobolev spaces. Leveraging this result, SATL enables the estimators to provably and universally adapt to the varying and unknown Sobolev smoothness of the source and offset functions. We derive the minimax lower bound of the learning problem in excess risk and show that SATL achieves a matching upper bound up to logarithmic factors. The optimal statistical rate reveals the factors influencing the transfer dynamics and efficacy, including the source sample size and the relative strength between domains. The theoretical findings and the effectiveness of SATL are confirmed by several experiments.

GNNs rely on the exchange of messages to distribute information along the edges of the graph. This approach makes the efficiency of architectures highly dependent on the specific structure of the input graph. Certain graph topologies lead to inefficient information propagation, resulting in a phenomenon known as over-squashing. While the majority of existing methods address over-squashing by rewiring the input graph, our novel approach involves constructing a graph directly from features using Delaunay Triangulation. We posit that the topological properties of the resulting graph prove advantageous for mitigate oversmoothing and over-squashing. Our extensive experimentation demonstrates that our method consistently outperforms established graph rewiring methods.

Reinforcement learning from human feedback (RLHF) has proven effective in aligning large language models (LLMs) with human preferences, but gathering high-quality preference labels is expensive. RL from AI Feedback (RLAIF), introduced in Bai et al. (2022b), offers a promising alternative that trains the reward model (RM) on preferences generated by an off-the-shelf LLM. Across the tasks of summarization, helpful dialogue generation, and harmless dialogue generation, we show that RLAIF achieves comparable performance to RLHF. Furthermore, we take a step towards "self-improvement" by demonstrating that RLAIF can outperform a supervised fine-tuned baseline even when the AI labeler is the same size as the policy, or even the exact same checkpoint as the initial policy. Finally, we introduce direct-RLAIF (d-RLAIF) - a technique that circumvents RM training by obtaining rewards directly from an off-the-shelf LLM during RL, which achieves superior performance to canonical RLAIF. Our results suggest that RLAIF can achieve performance on-par with using human feedback, offering a potential solution to the scalability limitations of RLHF.

It is well known that selecting samples with large losses/gradients can significantly reduce the number of training steps. However, the selection overhead is often too high to yield any meaningful gains in terms of overall training time. In this work, we focus on the greedy approach of selecting samples with large *approximate losses* instead of exact losses in order to reduce the selection overhead. For smooth convex losses, we show that such a greedy strategy can converge to a constant factor of the minimum value of the average loss in fewer iterations than the standard approach of random selection. We also theoretically quantify the effect of the approximation level. We then develop SIFT which uses early exiting to obtain approximate losses with an intermediate layer's representations for sample selection. We evaluate SIFT on the task of training a 110M parameter 12 layer BERT base model, and show significant gains (in terms of training hours and number of backpropagation steps) without any optimized implementation over vanilla training. For e.g., to reach 64% validation accuracy, SIFT with exit at the first layer takes $\sim$ 43 hours compared to $\sim$ 57 hours of vanilla training.

Despite large neural networks demonstrating remarkable abilities to complete different tasks, they require excessive memory usage to store the optimization states for training. To alleviate this, the low-rank adaptation (LoRA) is proposed to reduce the optimization states by training fewer parameters. However, LoRA restricts overall weight update matrices to be low-rank, limiting the model performance. In this work, we investigate the dynamics of LoRA and identify that it can be approximated by a random projection. Based on this observation, we propose Flora, which is able to achieve high-rank updates by resampling the projection matrices while enjoying the sublinear space complexity of optimization states. We conduct experiments across different tasks and model architectures to verify the effectiveness of our approach.

Agents built with large language models (LLMs) have shown great potential across a wide range of domains. However, in complex decision-making tasks, pure LLM-based agents tend to exhibit intrinsic bias in their choice of actions, which is inherited from the model's training data and results in suboptimal performance. To develop *strategic language agents*, i.e., agents that generate flexible language actions and possess strong decision-making abilities, we propose a novel framework that powers LLM-based agents with reinforcement learning (RL). We consider Werewolf, a popular social deduction game, as a challenging testbed that emphasizes versatile communication and strategic gameplay. To mitigate the intrinsic bias in language actions, our agents use an LLM to perform deductive reasoning and generate a diverse set of action candidates. Then an RL policy trained to optimize the decision-making ability chooses an action from the candidates to play in the game. Extensive experiments show that our agents overcome the intrinsic bias and outperform existing LLM-based agents in the Werewolf game. We also conduct human-agent experiments and find that our agents achieve human-level performance and demonstrate strong strategic play.

Studying the complex interactions between different brain regions is crucial in neuroscience. Various statistical methods have explored the latent communication across multiple brain regions. Two main categories are the Gaussian Process (GP) and Linear Dynamical System (LDS), each with unique strengths. The GP-based approach effectively discovers latent variables with frequency bands and communication directions. Conversely, the LDS-based approach is computationally efficient but lacks powerful expressiveness in latent representation. In this study, we merge both methodologies by creating an LDS mirroring a multi-output GP, termed Multi-Region Markovian Gaussian Process (MRM-GP). Our work establishes a connection between an LDS and a multi-output GP that explicitly models frequencies and phase delays within the latent space of neural recordings. Consequently, the model achieves a linear inference cost over time points and provides an interpretable low-dimensional representation, revealing communication directions across brain regions and separating oscillatory communications into different frequency bands.

Hybrid dynamical systems are prevalent in science and engineering to express complex systems with continuous and discrete states. To learn laws of systems, all previous methods for equation discovery in hybrid systems follow a two-stage paradigm, i.e. they first group time series into small cluster fragments and then discover equations in each fragment separately through methods in non-hybrid systems. Although effective, performance is then limited because these methods ignore the commonalities in the shared dynamics of fragments that are driven by the same equations. Besides, the two-stage paradigm breaks the interdependence between categorizing and representing dynamics that jointly form hybrid systems. In this paper, we reformulate the problem and propose an end-to-end learning framework, i.e. Amortized Equation Discovery (AMORE), to jointly categorize modes and discover equations characterizing motion dynamics of each mode by all segments of the mode. Experiments on four hybrid and six non-hybrid systems demonstrate the superior performance of our method against previous methods on equation discovery, segmentation, and forecasting.

Concept-based interpretability methods offer a lens into the internals of foundation models by decomposing their embeddings into high-level concepts. These concept representations are most useful when they are *compositional*, meaning that the individual concepts compose to explain the full sample. We show that existing unsupervised concept extraction methods find concepts which are not compositional. To automatically discover compositional concept representations, we identify two salient properties of such representations, and propose Compositional Concept Extraction (CCE) for finding concepts which obey these properties. We evaluate CCE on five different datasets over image and text data. Our evaluation shows that CCE finds more compositional concept representations than baselines and yields better accuracy on four downstream classification tasks.

We study how representation learning can improve the learning efficiency of contextual bandit problems. We study the setting where we play T linear contextual bandits with dimension simultaneously, and these T bandit tasks collectively share a common linear representation with a dimensionality of r ≪ d. We present a new algorithm based on alternating projected gradient descent (GD) and minimization estimator to recover a low-rank feature matrix. We obtain constructive provable guarantees for our estimator that provide a lower bound on the required sample complexity and an upper bound on the iteration complexity (total number of iterations needed to achieve a certain error level). Using the proposed estimator, we present a multi-task learning algorithm for linear contextual bandits and prove the regret bound of our algorithm. We presented experiments and compared the performance of our algorithm against benchmark algorithms.

Autonomous agents trained using deep reinforcement learning (RL) often lack the ability to successfully generalise to new environments, even when these environments share characteristics with the ones they have encountered during training. In this work, we investigate how the sampling of individual environment instances, or levels, affects the zero-shot generalisation (ZSG) ability of RL agents. We discover that, for deep actor-critic architectures sharing their base layers, prioritising levels according to their value loss minimises the mutual information between the agent's internal representation and the set of training levels in the generated training data. This provides a novel theoretical justification for the regularisation achieved by certain adaptive sampling strategies. We then turn our attention to unsupervised environment design (UED) methods, which assume control over level generation. We find that existing UED methods can significantly shift the training distribution, which translates to low ZSG performance. To prevent both overfitting and distributional shift, we introduce *data-regularised environment design* (DRED). DRED generates levels using a generative model trained to approximate the ground truth distribution of an initial set of level parameters. Through its grounding, DRED achieves significant improvements in ZSG over adaptive level sampling strategies and UED methods.